etd@IISc Collection:
http://hdl.handle.net/2005/18
2018-01-24T10:13:56ZPower Issues in SoCs : Power Aware DFT Architecture and Power Estimation
http://hdl.handle.net/2005/3003
Title: Power Issues in SoCs : Power Aware DFT Architecture and Power Estimation
Authors: Tudu, Jaynarayan Thakurdas
Abstract: Test power, data volume, and test time have been long-standing problems for sequential scan based testing of system-on-chip (SoC) design. The modern SoCs fabricated at lower technology nodes are complex in nature, the transistor count is as large as billions of gate for some of the microprocessors. The design complexity is further projected to increase in the coming years in accordance with Moore's law. The larger gate count and integration of multiple functionalities are the causes for higher test power dissipation, test time and data volume. The dynamic power dissipation during scan testing, i.e. during scan shift, launch and response capture, are major concerns for reliable as well as cost effective testing. Excessive average power dissipation leads to a thermal problem which causes burn-out of the chip during testing. Peak power on other hand causes test failure due to power induced additional delay. The test failure has direct impact on yield. The test power problem in modern 3D stacked based IC is even a more serious issue. Estimating the worst case functional power dissipation is yet another great challenge. The worst case functional power estimation is necessary because it gives an upper bound on the functional power dissipation which can further be used to determine the safe power zone for the test.
Several solutions in the past have been proposed to address these issues. In this thesis we have three major contributions: 1) Sequential scan chain reordering, and 2) JScan-an alternative Joint-scan DFT architecture to address primarily the test power issues along with test time and data volume, and 3) an integer linear programming methodology to address the power estimation problem. In order to reduce test power during shift, we have proposed a graph theoretic formulation for scan chain reordering and for optimum scan shift operation. For each formulation a set of algorithms is proposed. The experimental results on ISCAS-89 benchmark circuit show a reduction of around 25% and 15% in peak power and scan shift time respectively.
In order to have a holistic DFT architecture which could solve test power, test time, and data volume problems, a new DFT architecture called Joint-scan (JScan) have been developed. In JScan we have integrated the serial and random access scan architectures in a systematic way by which the JScan could harness the respective advantages from each of the architectures. The serial scan architecture
from test power, test time, and data volume problems. However, the serial scan is simple in terms of its functionality and is cost effective in terms of DFT circuitry. Whereas, the random ac-cess scan architecture is opposite to this; it is power efficient and it takes lesser time and data volume compared to serial scan. However, the random access scan occupies larger DFT area and introduces routing congestion. Therefore, we have proposed a methodology to realize the JScan architecture as an efficient alternative for standard serial and random access scan. Further, the JScan architecture is optimized and it resulted into a 2-Mode 2M-Jscan Joint-scan architecture. The proposed architectures are experimentally verified on larger benchmark circuits and compared with existing state of the art DFT architectures. The results show a reduction of 50% to 80% in test power and 30% to 50% in test time and data volume. The proposed architectures are also evaluated for routing area minimization and we obtained a saving of around 7% to 15% of chip area.
Estimating the worst case functional power being a challenging problem, we have proposed a binary integer linear programming (BILP) based methodology. Two different formulations have been proposed considering the different delay models namely zero-delay and unit-delay. The proposed methodology generates a pair or input vectors which could toggle the circuit to dissipate worst power. The BILP problems are solved using CPLEX solver for ISCAS-85 combinational benchmark circuits. For some of the circuits, the proposed methodology provided the worst possible power dissipation i.e. 80 to 100% toggling in nets.2018-01-09T18:30:00ZSymmetry in Scalar Fields
http://hdl.handle.net/2005/2989
Title: Symmetry in Scalar Fields
Authors: Thomas, Dilip Mathew
Abstract: Scalar fields are used to represent physical quantities measured over a domain of interest. Study of symmetric or repeating patterns in scalar fields is important in scientific data analysis because it gives deep insights into the properties of the underlying phenomenon.
This thesis proposes three methods to detect symmetry in scalar fields. The first
method models symmetry detection as a subtree matching problem in the contour tree, which is a topological graph abstraction of the scalar field. The contour tree induces a hierarchical segmentation of features at different scales and hence this method can detect symmetry at different scales. The second method identifies symmetry by comparing distances between extrema from each symmetric region. The distance is computed robustly using a topological abstraction called the extremum graph. Hence, this method can detect symmetry even in the presence of significant noise. The above methods compare
pairs of regions to identify symmetry instead of grouping the entire set of symmetric regions as a cluster. This motivates the third method which uses a clustering analysis for symmetry detection. In this method, the contours of a scalar field are mapped to points in a high-dimensional descriptor space such that points corresponding to similar contours lie in close proximity to each other. Symmetry is identified by clustering the points in the descriptor space.
We show through experiments on real world data sets that these methods are robust in
the presence of noise and can detect symmetry under different types of transformations. Extraction of symmetry information helps users in visualization and data analysis. We design novel applications that use symmetry information to enhance visualization of scalar field data and to facilitate their exploration.2018-01-08T18:30:00ZGame-Theoretic Analysis of Strategic Behaviour in Networks, Crowds and Classrooms
http://hdl.handle.net/2005/2955
Title: Game-Theoretic Analysis of Strategic Behaviour in Networks, Crowds and Classrooms
Authors: Vallam, Rohith Dwarakanath
Abstract: Over the past decade, the explosive growth of the Internet has led to a surge of interest to understand and predict aggregate behavior of large number of people or agents, particularly when they are connected through an underlying network structure. Numerous Internet-based applications have emerged that are as diverse as getting micro-tasks executed through online labor markets (also known as crowd sourcing) to acquiring new skills through massively open online courses (also known as MOOCs). However, there has been a major inadequacy in existing studies with respect to evaluating the impact of strategic behavior of the agents participating in such networks, crowds, and classrooms. The primary focus of this doctoral work is to understand the equilibrium behaviour emerging from these real-world, strategic environments by blending ideas from the areas of game theory, graph theory, and optimization, to derive novel solutions to these new-age economic models. In particular, we investigate the following three research challenges:
(1) How do strategic agents form connections with one another? Will it ever happen that strategically stable networks are social welfare maximizing as well?
(2) How do we design mechanisms for eliciting truthful feedback about an object (perhaps a new product or service or person) from a crowd of strategic raters? What can we tell about these mechanisms when the raters are connected through a social network?
(3) How do we incentivize better participation of instructors and students in online edu-cation forums? Can we recommend optimal strategies to students and instructors to get the best out of these forums?2018-01-02T18:30:00ZConsistency of Spectral Algorithms for Hypergraphs under Planted Partition Model
http://hdl.handle.net/2005/2947
Title: Consistency of Spectral Algorithms for Hypergraphs under Planted Partition Model
Authors: Ghoshdastidar, Debarghya
Abstract: Hypergraph partitioning lies at the heart of a number of problems in machine learning as well as other engineering disciplines. While partitioning uniform hypergraphs is often required in computer vision problems that involve multi-way similarities, non-uniform hypergraph partitioning has applications in database systems, circuit design etc. As in the case of graphs, it is known that for given objective and balance constraints, the problem of optimally partitioning a hypergraph is NP-hard. Yet, over the last two decades, several efficient heuristics have been studied in the literature and their empirical success is widely appreciated. In contrast to the extensive studies related to graph partitioning, the theoretical guarantees of hypergraph partitioning approaches have not received much attention in the literature. The purpose of this thesis is to establish the statistical error bounds for certain spectral algorithms for partitioning uniform as well as non-uniform hypergraphs.
The mathematical framework considered in this thesis is the following. Let V be a set of n vertices, and ψ : V ->{1,…,k} be a (hidden) partition of V into k classes. A random hypergraph (V,E) is generated according to a planted partition model, i.e., subsets of V are independently added to the edge set E with probabilities depending on the class memberships of the participating vertices. Let ψ' be the partition of V obtained from a certain algorithm acting on a random realization of the hypergraph. We provide an upper bound on the number of disagreements between ψ and ψ'. To be precise, we show that under certain conditions, the asymptotic error is o(n) with probability (1-o(1)). In the existing literature, such error rates are only known in the case of graphs (Rohe et al., Ann. Statist., 2011; Lei \& Rinaldo, Ann. Statist., 2015), where the planted model coincides with the popular stochastic block model. Our results are based on matrix concentration inequalities and perturbation bounds, and the derived bounds can be used to comment on the consistency of spectral hypergraph partitioning algorithms.
It is quite common in the literature to resort to a spectral approach when the quantity of interest is a matrix, for instance, the adjacency or Laplacian matrix for graph partitioning. This is certainly not true for hypergraph partitioning as the adjacency relations cannot be encoded into a symmetric matrix as in the case of graphs. However, if one restricts the problem to m-uniform hypergraphs for some m ≥ 2, then a symmetric tensor of order m can be used to express the multi-way interactions or adjacencies. Thus, the use of tensor spectral algorithms, based on the spectral theory of symmetric tensors, is a natural choice in this scenario. We observe that a wide variety of uniform hypergraph partitioning methods studied in the literature can be related to any one of two principle approaches: (1) solving a tensor trace maximization problem, or (2) use of the higher order singular value decomposition of tensors. We derive statistical error bounds to show that both these approaches lead to consistent partitioning algorithms.
Our results also hold when the hypergraph under consideration allows weighted edges, a situation that is commonly encountered in computer vision applications such as motion segmentation, image registration etc. In spite of the theoretical guarantees, a tensor spectral approach is not preferable in this setting due to the time and space complexity of computing the weighted adjacency tensor. Keeping this practical scenario in mind, we prove that consistency can still be achieved by incorporating certain tensor sampling strategies. In particular, we show that if the edges are sampled according to certain distribution, then consistent partitioning can be achieved with only few sampled edges. Experiments on benchmark problems demonstrate that such sampled tensor spectral algorithms are indeed useful in practice.
While vision tasks mostly involve uniform hypergraphs, in database and electronics applications, one often finds non-uniform hypergraphs with edges of varying sizes. These hypergraphs cannot be expressed in terms of adjacency matrices or tensors, and hence, use of a spectral approach is tricky in this context. The partitioning problem gets more challenging due to the fact that, in practice, these hypergraphs are quite sparse, and hence, provide less information about the partition. We consider spectral algorithms for partitioning clique and star expansions of hypergraphs, and study their consistency under a sparse planted partition model.
The results of hypergraph partitioning can be further extended to address the well-known hypergraph vertex coloring problem, where the objective is to color the vertices such that no edge is monochromatic. The hardness of this problem is well established. In fact, even when a hypergraph is bipartite or 2-colorable, it is NP-hard to find a proper 2-coloring for it. We propose a spectral coloring algorithm, and show that if the non-monochromatic subsets of vertices are independently added to the edge set with certain probabilities, then with probability (1-o(1)), our algorithm succeeds in coloring bipartite hypergraphs with only two colors.
To the best our knowledge, these are the first known results related to consistency of partitioning general hypergraphs.2017-12-31T18:30:00Z